No abstract

Traffic analyses from various packet networks have shown the self-similar nature of the bursty network traffic. Among the existing self-similar traffic models the Fractional Brownian Motion (FBM) process has been regarded by some works as an attractive alternative to traditional modeling approaches despite its limitations. The parsimonious FBM traffic modeling involves three parameters: the mean rate (m), the variance coefficient (a) and the Hurst parameter (H). Fast and correct estimation of these parameters from measured data is very important for performance studies and other network dimensioning work. This paper has two goals. First, the existing wavelet-based estimator is extended to do on-line estimation of the three FBM traffic parameters. Second, with the help of the on-line estimation tool, we perform extensive simulation experiments under the consideration of higher-layer protocol actions (TCP connections). We verify the FBM model for two general characteristics of user actions/behaviors that today appear to be of great significance: (a) heavy-tailed file sizes, and (b) heavy-tailed distributions of connection inter-arrival times. Our simulation results show that the FBM model works well for middle to large link loads. For small or too large loads, the FBM model significantly underestimates or overestimates the queuing performance.